面向中文的事件抽取的研究与实现.doc

中国最大的论文知识平台 事件抽取是信息抽取领域的一个重要研究方向。事件抽取主要把人们用自然语言表达的事件,以结构化的形式表现出来。事件抽取作为信息处理领域的关键技术,在信息检索、自动问答、自动摘要、数据挖掘、文本挖掘等领域有着广泛的应用。本文主要针对中文事件抽取的两大主要任务:事件类别的识别和事件元素的识别进行了学****与探索,并实现了一个简易的中文事件抽取程序。在事件类别识别中,本文采用基于触发词来识别事件类别的方法。首先根据训练语料获取触发词和其对应的候选类别建立字典。然后根据字典和训练语料识别出文档中的触发词和其对应的候选事件类别,以此确定候选事件。再通过使用词法,上下文信息和事件模板等特征,利用分类器对候选事件进行二元分类,从候选事件中提取出真正的事件。在事件元素识别中,本文采用基于事件模板,通过特征提取进行多元分类的方法来识别事件元素。首先通过事件类别可以获得事件模板,事件模板是由事件角色组成的。将事件元素识别看成分类问题,再通过词法,上下文信息和类别信息等特征描述候选元素,针对每一类事件构建分类器进行多元分类,提取出符合要求的事件元素。本文设计并实现了一个简单的中文事件抽取系统,经测试最终结果为 % 。关键词: 事件抽取;事件类别识别;事件元素识别中国最大的论文知识平台 tract Event extraction is an important area of information extraction research. Event extraction explains the events which people use natural language to express with a structured form. Event extraction as the key technology of information processing, has been widely used in information retrieval, question answering, automatic summarization, data mining, text mining and other fields. In this paper, we pay attention to two main tasks: Event types recognition and event element recognition. According to this, we go to learn and exploration, e true a simple extraction procedure of Chinese events. Recognition of the event types, we use the trigger-word-based approach to recognition event type. First of all, according to training data we can get trigger words and their candidate type, building a dictionary. Secondly , according to the dictionary and training data we can recognition the trigger words and the event type ofa document. We use this to confirm the candidate events. Thirdly, we use classification of binary to classify the candidate events, by lexical, context information and event templates and other features. Recognition of the event element, we use event-template-based way to feature extraction for multi-classification methods to recognition the event element. First , using event types we can get the event templates, and we can also gain the elements of the candidate events from the event templates. The templates are